1 /* 2 * Copyright (c) 1999, 2023, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2014, 2021, Red Hat Inc. All rights reserved. 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This code is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 only, as 8 * published by the Free Software Foundation. 9 * 10 * This code is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 13 * version 2 for more details (a copy is included in the LICENSE file that 14 * accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License version 17 * 2 along with this work; if not, write to the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 21 * or visit www.oracle.com if you need additional information or have any 22 * questions. 23 * 24 */ 25 26 #include "precompiled.hpp" 27 #include "c1/c1_MacroAssembler.hpp" 28 #include "c1/c1_Runtime1.hpp" 29 #include "gc/shared/barrierSetAssembler.hpp" 30 #include "gc/shared/collectedHeap.hpp" 31 #include "gc/shared/tlab_globals.hpp" 32 #include "interpreter/interpreter.hpp" 33 #include "oops/arrayOop.hpp" 34 #include "oops/markWord.hpp" 35 #include "runtime/basicLock.hpp" 36 #include "runtime/os.hpp" 37 #include "runtime/sharedRuntime.hpp" 38 #include "runtime/stubRoutines.hpp" 39 40 void C1_MacroAssembler::float_cmp(bool is_float, int unordered_result, 41 FloatRegister f0, FloatRegister f1, 42 Register result) 43 { 44 Label done; 45 if (is_float) { 46 fcmps(f0, f1); 47 } else { 48 fcmpd(f0, f1); 49 } 50 if (unordered_result < 0) { 51 // we want -1 for unordered or less than, 0 for equal and 1 for 52 // greater than. 53 cset(result, NE); // Not equal or unordered 54 cneg(result, result, LT); // Less than or unordered 55 } else { 56 // we want -1 for less than, 0 for equal and 1 for unordered or 57 // greater than. 58 cset(result, NE); // Not equal or unordered 59 cneg(result, result, LO); // Less than 60 } 61 } 62 63 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) { 64 const int aligned_mask = BytesPerWord -1; 65 const int hdr_offset = oopDesc::mark_offset_in_bytes(); 66 assert_different_registers(hdr, obj, disp_hdr); 67 int null_check_offset = -1; 68 69 verify_oop(obj); 70 71 // save object being locked into the BasicObjectLock 72 str(obj, Address(disp_hdr, BasicObjectLock::obj_offset())); 73 74 null_check_offset = offset(); 75 76 if (DiagnoseSyncOnValueBasedClasses != 0) { 77 load_klass(hdr, obj); 78 ldrw(hdr, Address(hdr, Klass::access_flags_offset())); 79 tstw(hdr, JVM_ACC_IS_VALUE_BASED_CLASS); 80 br(Assembler::NE, slow_case); 81 } 82 83 // Load object header 84 ldr(hdr, Address(obj, hdr_offset)); 85 if (LockingMode == LM_LIGHTWEIGHT) { 86 fast_lock(obj, hdr, rscratch1, rscratch2, slow_case); 87 } else if (LockingMode == LM_LEGACY) { 88 Label done; 89 // and mark it as unlocked 90 orr(hdr, hdr, markWord::unlocked_value); 91 // save unlocked object header into the displaced header location on the stack 92 str(hdr, Address(disp_hdr, 0)); 93 // test if object header is still the same (i.e. unlocked), and if so, store the 94 // displaced header address in the object header - if it is not the same, get the 95 // object header instead 96 lea(rscratch2, Address(obj, hdr_offset)); 97 cmpxchgptr(hdr, disp_hdr, rscratch2, rscratch1, done, /*fallthough*/nullptr); 98 // if the object header was the same, we're done 99 // if the object header was not the same, it is now in the hdr register 100 // => test if it is a stack pointer into the same stack (recursive locking), i.e.: 101 // 102 // 1) (hdr & aligned_mask) == 0 103 // 2) sp <= hdr 104 // 3) hdr <= sp + page_size 105 // 106 // these 3 tests can be done by evaluating the following expression: 107 // 108 // (hdr - sp) & (aligned_mask - page_size) 109 // 110 // assuming both the stack pointer and page_size have their least 111 // significant 2 bits cleared and page_size is a power of 2 112 mov(rscratch1, sp); 113 sub(hdr, hdr, rscratch1); 114 ands(hdr, hdr, aligned_mask - (int)os::vm_page_size()); 115 // for recursive locking, the result is zero => save it in the displaced header 116 // location (null in the displaced hdr location indicates recursive locking) 117 str(hdr, Address(disp_hdr, 0)); 118 // otherwise we don't care about the result and handle locking via runtime call 119 cbnz(hdr, slow_case); 120 // done 121 bind(done); 122 } 123 increment(Address(rthread, JavaThread::held_monitor_count_offset())); 124 return null_check_offset; 125 } 126 127 128 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) { 129 const int aligned_mask = BytesPerWord -1; 130 const int hdr_offset = oopDesc::mark_offset_in_bytes(); 131 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different"); 132 Label done; 133 134 if (LockingMode != LM_LIGHTWEIGHT) { 135 // load displaced header 136 ldr(hdr, Address(disp_hdr, 0)); 137 // if the loaded hdr is null we had recursive locking 138 // if we had recursive locking, we are done 139 cbz(hdr, done); 140 } 141 142 // load object 143 ldr(obj, Address(disp_hdr, BasicObjectLock::obj_offset())); 144 verify_oop(obj); 145 146 if (LockingMode == LM_LIGHTWEIGHT) { 147 ldr(hdr, Address(obj, oopDesc::mark_offset_in_bytes())); 148 // We cannot use tbnz here, the target might be too far away and cannot 149 // be encoded. 150 tst(hdr, markWord::monitor_value); 151 br(Assembler::NE, slow_case); 152 fast_unlock(obj, hdr, rscratch1, rscratch2, slow_case); 153 } else if (LockingMode == LM_LEGACY) { 154 // test if object header is pointing to the displaced header, and if so, restore 155 // the displaced header in the object - if the object header is not pointing to 156 // the displaced header, get the object header instead 157 // if the object header was not pointing to the displaced header, 158 // we do unlocking via runtime call 159 if (hdr_offset) { 160 lea(rscratch1, Address(obj, hdr_offset)); 161 cmpxchgptr(disp_hdr, hdr, rscratch1, rscratch2, done, &slow_case); 162 } else { 163 cmpxchgptr(disp_hdr, hdr, obj, rscratch2, done, &slow_case); 164 } 165 // done 166 bind(done); 167 } 168 decrement(Address(rthread, JavaThread::held_monitor_count_offset())); 169 } 170 171 172 // Defines obj, preserves var_size_in_bytes 173 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) { 174 if (UseTLAB) { 175 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case); 176 } else { 177 b(slow_case); 178 } 179 } 180 181 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) { 182 assert_different_registers(obj, klass, len); 183 if (UseCompactObjectHeaders) { 184 ldr(t1, Address(klass, Klass::prototype_header_offset())); 185 str(t1, Address(obj, oopDesc::mark_offset_in_bytes())); 186 } else { 187 // This assumes that all prototype bits fit in an int32_t 188 mov(t1, (int32_t)(intptr_t)markWord::prototype().value()); 189 str(t1, Address(obj, oopDesc::mark_offset_in_bytes())); 190 191 if (UseCompressedClassPointers) { // Take care not to kill klass 192 encode_klass_not_null(t1, klass); 193 strw(t1, Address(obj, oopDesc::klass_offset_in_bytes())); 194 } else { 195 str(klass, Address(obj, oopDesc::klass_offset_in_bytes())); 196 } 197 } 198 199 if (len->is_valid()) { 200 strw(len, Address(obj, arrayOopDesc::length_offset_in_bytes())); 201 } else if (UseCompressedClassPointers && !UseCompactObjectHeaders) { 202 store_klass_gap(obj, zr); 203 } 204 } 205 206 // preserves obj, destroys len_in_bytes 207 // 208 // Scratch registers: t1 = r10, t2 = r11 209 // 210 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1, Register t2) { 211 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0"); 212 assert(t1 == r10 && t2 == r11, "must be"); 213 214 Label done; 215 216 // len_in_bytes is positive and ptr sized 217 subs(len_in_bytes, len_in_bytes, hdr_size_in_bytes); 218 br(Assembler::EQ, done); 219 220 // zero_words() takes ptr in r10 and count in words in r11 221 mov(rscratch1, len_in_bytes); 222 lea(t1, Address(obj, hdr_size_in_bytes)); 223 lsr(t2, rscratch1, LogBytesPerWord); 224 address tpc = zero_words(t1, t2); 225 226 bind(done); 227 if (tpc == nullptr) { 228 Compilation::current()->bailout("no space for trampoline stub"); 229 } 230 } 231 232 233 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) { 234 assert_different_registers(obj, t1, t2); // XXX really? 235 assert(header_size >= 0 && object_size >= header_size, "illegal sizes"); 236 237 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case); 238 239 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB); 240 } 241 242 // Scratch registers: t1 = r10, t2 = r11 243 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, bool is_tlab_allocated) { 244 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, 245 "con_size_in_bytes is not multiple of alignment"); 246 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize; 247 248 initialize_header(obj, klass, noreg, t1, t2); 249 250 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) { 251 // clear rest of allocated space 252 const Register index = t2; 253 if (var_size_in_bytes != noreg) { 254 mov(index, var_size_in_bytes); 255 initialize_body(obj, index, hdr_size_in_bytes, t1, t2); 256 if (Compilation::current()->bailed_out()) { 257 return; 258 } 259 } else if (con_size_in_bytes > hdr_size_in_bytes) { 260 con_size_in_bytes -= hdr_size_in_bytes; 261 lea(t1, Address(obj, hdr_size_in_bytes)); 262 address tpc = zero_words(t1, con_size_in_bytes / BytesPerWord); 263 if (tpc == nullptr) { 264 Compilation::current()->bailout("no space for trampoline stub"); 265 return; 266 } 267 } 268 } 269 270 membar(StoreStore); 271 272 if (CURRENT_ENV->dtrace_alloc_probes()) { 273 assert(obj == r0, "must be"); 274 far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id))); 275 } 276 277 verify_oop(obj); 278 } 279 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int base_offset_in_bytes, int f, Register klass, Label& slow_case) { 280 assert_different_registers(obj, len, t1, t2, klass); 281 282 // determine alignment mask 283 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work"); 284 285 // check for negative or excessive length 286 mov(rscratch1, (int32_t)max_array_allocation_length); 287 cmp(len, rscratch1); 288 br(Assembler::HS, slow_case); 289 290 const Register arr_size = t2; // okay to be the same 291 // align object end 292 mov(arr_size, (int32_t)base_offset_in_bytes + MinObjAlignmentInBytesMask); 293 add(arr_size, arr_size, len, ext::uxtw, f); 294 andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask); 295 296 try_allocate(obj, arr_size, 0, t1, t2, slow_case); 297 298 initialize_header(obj, klass, len, t1, t2); 299 300 // Clear leading 4 bytes, if necessary. 301 // TODO: This could perhaps go into initialize_body() and also clear the leading 4 bytes 302 // for non-array objects, thereby replacing the klass-gap clearing code in initialize_header(). 303 int base_offset = base_offset_in_bytes; 304 if (!is_aligned(base_offset, BytesPerWord)) { 305 assert(is_aligned(base_offset, BytesPerInt), "must be 4-byte aligned"); 306 strw(zr, Address(obj, base_offset)); 307 base_offset += BytesPerInt; 308 } 309 assert(is_aligned(base_offset, BytesPerWord), "must be word-aligned"); 310 311 // clear rest of allocated space 312 initialize_body(obj, arr_size, base_offset, t1, t2); 313 if (Compilation::current()->bailed_out()) { 314 return; 315 } 316 317 membar(StoreStore); 318 319 if (CURRENT_ENV->dtrace_alloc_probes()) { 320 assert(obj == r0, "must be"); 321 far_call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id))); 322 } 323 324 verify_oop(obj); 325 } 326 327 328 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) { 329 verify_oop(receiver); 330 // explicit null check not needed since load from [klass_offset] causes a trap 331 // check against inline cache. This is checked in Universe::genesis().. 332 cmp_klass(receiver, iCache, rscratch1); 333 } 334 335 336 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes) { 337 assert(bang_size_in_bytes >= framesize, "stack bang size incorrect"); 338 // Make sure there is enough stack space for this method's activation. 339 // Note that we do this before creating a frame. 340 generate_stack_overflow_check(bang_size_in_bytes); 341 MacroAssembler::build_frame(framesize); 342 343 // Insert nmethod entry barrier into frame. 344 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler(); 345 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */, nullptr /* guard */); 346 } 347 348 void C1_MacroAssembler::remove_frame(int framesize) { 349 MacroAssembler::remove_frame(framesize); 350 } 351 352 353 void C1_MacroAssembler::verified_entry(bool breakAtEntry) { 354 // If we have to make this method not-entrant we'll overwrite its 355 // first instruction with a jump. For this action to be legal we 356 // must ensure that this first instruction is a B, BL, NOP, BKPT, 357 // SVC, HVC, or SMC. Make it a NOP. 358 nop(); 359 } 360 361 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) { 362 // rfp, + 0: link 363 // + 1: return address 364 // + 2: argument with offset 0 365 // + 3: argument with offset 1 366 // + 4: ... 367 368 ldr(reg, Address(rfp, (offset_in_words + 2) * BytesPerWord)); 369 } 370 371 #ifndef PRODUCT 372 373 void C1_MacroAssembler::verify_stack_oop(int stack_offset) { 374 if (!VerifyOops) return; 375 verify_oop_addr(Address(sp, stack_offset)); 376 } 377 378 void C1_MacroAssembler::verify_not_null_oop(Register r) { 379 if (!VerifyOops) return; 380 Label not_null; 381 cbnz(r, not_null); 382 stop("non-null oop required"); 383 bind(not_null); 384 verify_oop(r); 385 } 386 387 void C1_MacroAssembler::invalidate_registers(bool inv_r0, bool inv_r19, bool inv_r2, bool inv_r3, bool inv_r4, bool inv_r5) { 388 #ifdef ASSERT 389 static int nn; 390 if (inv_r0) mov(r0, 0xDEAD); 391 if (inv_r19) mov(r19, 0xDEAD); 392 if (inv_r2) mov(r2, nn++); 393 if (inv_r3) mov(r3, 0xDEAD); 394 if (inv_r4) mov(r4, 0xDEAD); 395 if (inv_r5) mov(r5, 0xDEAD); 396 #endif 397 } 398 #endif // ifndef PRODUCT